To cheat or not to cheat : The dark side of data science

Objectifs

  • This meeting is not a lecture course, but an exchange session dedicated to unorthodox practices in the field of research and more particularly, in the field of data science, but not only.
  • If there is a (very) dark side of research practices, there are also subtle margins within it could be helpful to move forward. And this time, it’s not about tricking. It’s about being more understandable. When cheating is not cheating.

Talks are delivered in English.

Vidéos des interventions

Programme

The event is organized in partnership with the doctoral school “Mathematics and Computing”.

Talks are delivered in English.

This meeting is not a lecture course, but an exchange session dedicated to unorthodox practices in the field of research and more particularly, in the field of data science, but not only.

If there is a (very) dark side of research practices, there are also subtle margins within it could be helpful to move forward. And this time, it’s not about tricking. It’s about being more understandable. When cheating is not cheating.

The point of this workshop is to help you to know where you stand, to give you a compass in a tortured landscape.

Will you meet the challenge of orthodoxe research practices?

To start the fight, first, you should know your enemy : p-hacking, low statistical power, failure to control bias, poor quality control, falsified data, writing your own peer-review, but also working with predatory publishers, keeping your code veiled

Do you see patterns in random data (aka apophenia ; Munafò et al., 2017)? Do you have « the tendency to focus on evidence that is in line with our expectations or favoured explanation » (aka confirmation bias ; Munafò et al., 2017)? Or maybe have you heard of « the tendency to see an event as having been predictable only after it has occurred » (aka hindsight bias ; Munafò et al., 2017)?

But you also should be aware of the wide and sometimes surprising range of possibilities to produce more understandable results and thus, a better research. Get the grip, it’s up to you!

Nicolas Rougier : “Ten Simple Rules for Scientific Fraud & Misconduct”

Abstract : « We obviously do not encourage scientific fraud nor misconduct. The goal of this talk is to alert the audience to problems that have arisen in part due to the Publish or Perish imperative, which has driven a number of researchers to cross the Rubicon without the full appreciation of the consequences. Choosing fraud will hurt science, end careers, and could have impacts on life outside of the lab. If you’re tempted (even slightly) to beautify your results, keep in mind that the benefits are probably not worth the risks », (N. Rougier, J. Timmer 2017)

Nicolas Rougier est chercheur Inria en neurosciences computationelles travaillant à l’institut des maladies neurodégénératives à Bordeaux. Il a co-fondé le journal ReScience qui est spécialisé dans la publication de réplication en sciences computationelles.

Christophe Bontemps : « How To Lie With Graphics? »

Abstract : « According to Mark Twain « There are three kinds of lies: lies, damned lies, and statistics’’.  Today, with the emergence of so-called Data Science and self-proclaimed data scientists, we observe that graphical lies are everywhere.  They are even more powerful than spurious statistics. Many graphics in blogs, newspapers, and TV convey information that is misleading, by mistake or on purpose.  I propose a short tutorial to visual fallacies and lies. My goal here is not to encourage cheating and lying, but on the contrary to highlight the techniques used to elaborate misleading data visualizations.  This introduction should help researchers, citizens, (data) journalists and decision makers to distinguish visual lies from consistent graphics.  » Chr. Bontemps

Christophe Bontemps est ingénieur de recherche à l’INRA au sein de la Toulouse School of Economics. Economètre-Statisticien de formation, il enseigne la visualisation des données depuis plusieurs années et est co-organisateur du Meetup Toulouse Dataviz.

Sources

Munafò, Marcus R., Brian A. Nosek, Dorothy V. M. Bishop, Katherine S. Button, Christopher D. Chambers, Nathalie Percie du Sert, Uri Simonsohn, Eric-Jan Wagenmakers, Jennifer J. Ware, and John P. A. Ioannidis. 2017. ‘A Manifesto for Reproducible Science’. Nature Human Behaviour 1 (1): 0021. https://doi.org/10.1038/s41562-016-0021.

Rougier, Nicolas P., and John Timmer. 2017. “Ten Simple Rules for Scientific Fraud & Misconduct”. https://hal.inria.fr/hal-01562601.

Les commentaires sont fermés.